XLS-R-LUGANDA-ASR-CV-14-1B

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on the common_voice_14_0 dataset. It achieves the following results on the evaluation set:

• Loss: inf
• Wer: 0.3060
• Cer: 0.0713

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0003
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Validation Loss	Wer
1.5535	0.18	400	0.1685	inf	0.6590
0.539	0.36	800	0.1516	inf	0.5934
0.49	0.54	1200	0.1365	inf	0.5466
0.4569	0.72	1600	0.1364	inf	0.5523
0.4845	0.45	2000	0.1525	inf	0.5907
0.4592	0.54	2400	0.1485	inf	0.5766
0.4447	0.63	2800	0.1397	inf	0.5482
0.426	0.72	3200	0.1352	inf	0.5290
0.4454	0.81	3600	inf	0.5330	0.1333
0.4188	0.9	4000	inf	0.4903	0.1240
0.4083	0.99	4400	inf	0.4857	0.1226
0.367	1.08	4800	inf	0.4499	0.1114
0.3468	1.17	5200	inf	0.4345	0.1063
0.3401	1.27	5600	inf	0.4130	0.1009
0.3269	1.36	6000	inf	0.4113	0.1004
0.3171	1.45	6400	inf	0.3934	0.0956
0.2996	1.54	6800	inf	0.3803	0.0913
0.288	1.63	7200	inf	0.3681	0.0891
0.2812	1.72	7600	inf	0.3573	0.0853
0.2699	1.81	8000	inf	0.3504	0.0835
0.2584	1.9	8400	inf	0.3343	0.0786
0.2424	1.99	8800	inf	0.3232	0.0759
0.2201	2.08	9200	inf	0.3176	0.0740
0.2031	2.17	9600	inf	0.3085	0.0719
0.2007	2.26	10000	inf	0.3060	0.0713

Framework versions

• Transformers 4.38.1
• Pytorch 2.2.1
• Datasets 2.17.0
• Tokenizers 0.15.2